As far as I know, the concept of "democratizing AI" comes specifically from the OpenAI initiative, which intends to make new developments in AI technology broadly available and not tied to any single company or entity. This in order to ensure that no single entity has control over how this technology can be used.

Given that Google, Apple, Facebook and others have vastly more data than any independent project, and therefore have stronger AI (currently limited to e.g. speech recognition, image recognition and other low-impact applications), the state of democratization of AI by this measure is poor today.

If a CEO says the word 'democratisation' in relation to almost anything you can be guaranteed that the end result will have nothing to do with enhancing democracy.